Live example
Duolingo — Streak Repair
Following INI-001 from first signal to second iteration cycle.
A signal arrives — usage data, a customer complaint, a gut feeling. Before writing a single requirement, the problem is framed. What is actually broken? For whom? Under what conditions? This stage ends with a precise, falsifiable problem statement.
Problem-Framer
JTBD-Mapper
Persona-Builder
Assumption-Extractor
When-You-Have-a-New-Idea
Duolingo example
Users who break a streak churn at 3x the baseline rate. Wednesday is the highest streak-break day — and push open rates show they did open the app. They saw the broken streak and closed it. Not a reminder problem. A recovery problem.
Problem-Framer output → framed statement
Every initiative is a bet. This stage converts the framed problem into a falsifiable hypothesis and a minimum viable experiment. No build decision is made without a clear measure of what success looks like — and what would prove the bet wrong.
Hypothesis-Builder
Experiment-Designer
Opportunity-Tree
Edge-Case-Finder
Duolingo example
"If we offer a one-time streak repair option within 24 hours of a missed day, D30 retention for streak-break users will increase by at least 20% vs. the control group."
H-Hypotheses/HYP-001.md
The initiative is formally defined, linked to a Strategic Bet, and scored against competing work. Stakeholder dependencies are mapped. A business case is built for anything that needs leadership alignment. The initiative enters the roadmap only after this gate is cleared.
Decision-Architect
Priority-Stack
Tradeoff-Mapper
Business-Case-Builder
Cross-Team-Mapper
Duolingo example
INI-001 approved. Linked to Bet: D30 retention. Open question deferred: should repair cost Gems or be free? Ship free first — decide on monetization post-signal, not pre-signal.
I-Initiatives/INI-001.md
Sprint is planned against initiative capacity. A PRD is generated from the validated initiative. During development, the PM uses Build-Companion to handle in-sprint decisions, scope changes, and edge cases without losing structural discipline.
/sprint — Sprint-Planner
PRD-Builder
Build-Companion
Build-Review
When-Starting-a-Sprint
Duolingo example
Sprint goal: ship the streak repair prompt on iOS and Android. Free tier only. Scope locked. Push notification timing (midnight vs. next open) deferred — flagged in INI-001 for next cycle.
PRD generated → Sprint scope committed
3–5 days before go-live: the Launch-Coordinator runs a go/no-go checklist, surfaces any missing dependencies, and ensures stakeholders are briefed. Each stakeholder gets a version of the launch message matched to what they actually care about.
/launch — Launch-Coordinator
Stakeholder-Translator
Narrative-Refiner
Board-Update-Builder
When-Launching
Duolingo example
Go/no-go: cleared. Growth team briefed on retention messaging. Analytics confirmed D30 tracking and streak_repair_accepted events are wired. A/B test EXP-001 activated — 50% of streak-break users see the repair prompt, 50% see current empty state.
P-Proof/EXP-001.md → active
At the agreed measurement window, data is read against the hypothesis. The Performance-Tracker enforces interpretation discipline — a result without an interpretation is not a closed loop. The hypothesis is declared validated, invalidated, or inconclusive.
Performance-Tracker
/review-experiment
Sprint-Summarizer
When-Data-Comes-In
Duolingo example
Day-30 read. Repair-accepted users show 31% higher D30 retention vs. control — hypothesis validated. Dismissal rate 42% flagged as a secondary signal worth investigating next cycle.
H-Hypotheses/HYP-001.md → validated
The measurement result feeds directly into the next decision. If validated: double down and expand the bet. If invalidated: return to Discovery with stronger constraints. The loop closes, and compounds.
/iterate — Iteration-Planner
/weekly — Weekly-Review
Priority-Stack
Tradeoff-Mapper
When-Closing-the-Week
Duolingo example
Hypothesis validated. INI-002 scoped: test Gem-cost repair (monetization angle). Parallel open question from Stage 6: why is dismissal rate 42%? Edge-Case-Finder flagged prompt timing as a candidate — investigate next cycle.
I-Initiatives/INI-002.md → Defined
Tool types
Agent — persistent intelligence with memory and judgment
Skill — single-task specialist, runs once, returns structured output
Workflow — multi-step orchestration triggered by a slash command
Playbook — situation-based prompt that loads the right tools automatically